AI-generated (not entirely realistic) railway images

[popeter45]

my attempt at a 373 in TfW livery resulted in a early 2000's version of a IET that works surpisingly well

 

[signed]

my attempt at a 373 in TfW livery gave me resulted in a early 2000's version of a IET that works surpisingly well
View attachment 183600

I thought that was a ETR500 mashup , which would be period correct

 

[frankmoh]

Considering how all my previous prompt attempts to generate a networker have failed, can you just try "Class 465 train" I want to see which livery it puts out.

 

[signed]

Considering how all my previous prompt attempts to generate a networker have failed, can you just try "Class 465 train" I want to see which livery it puts out.

 

[frankmoh]

View attachment 183609

Skynet fails to comprehend the Kent Link, we're safe for now.

 

[Gloster]

Has anyone considered what this is going to do for photography? If I understand correctly AI has reached a state where the user can get a pretty good image and then gradually tweak it to produce an image without any obvious faults. How will we then know that, say, when someone claims to have photographed a Class ??? at X they are telling the truth. I know that there are plenty of ways of checking and many of the wilder or more brazen ones will be disproved (until someone fiddles with the sources of information), but some people are clever enough to fool the system. Whither or wither the truth?

Perhaps this should be a separate thread.

 

[PTR 444]

View attachment 183609

Looks like a Class 319/Class 455 hybrid with an alternate variation of post-NSE original SWT livery.

 

[6Gman]

View attachment 183562

Why's the driver in the secondman's seat?

 

[frankmoh]

Has anyone considered what this is going to do for photography? If I understand correctly AI has reached a state where the user can get a pretty good image and then gradually tweak it to produce an image without any obvious faults. How will we then know that, say, when someone claims to have photographed a Class ??? at X they are telling the truth. I know that there are plenty of ways of checking and many of the wilder or more brazen ones will be disproved (until someone fiddles with the sources of information), but some people are clever enough to fool the system. Whither or wither the truth?

Perhaps this should be a separate thread.

Well given how it seems to not understand what a 465 looks like we will be safe. Unless the fake trains are Sprinters or 455s.

 

[eldomtom2]

Has anyone considered what this is going to do for photography? If I understand correctly AI has reached a state where the user can get a pretty good image and then gradually tweak it to produce an image without any obvious faults. How will we then know that, say, when someone claims to have photographed a Class ??? at X they are telling the truth. I know that there are plenty of ways of checking and many of the wilder or more brazen ones will be disproved (until someone fiddles with the sources of information), but some people are clever enough to fool the system. Whither or wither the truth?

Perhaps this should be a separate thread.

That assumes that AI image generation will get to the point where obvious rail-related faults will be absent. I'm not so sure that will be the case - it's a field where everything has to be dead on and it's one where AI companies aren't going to put any specific attention in due to its niche nature.
Frankly anyone trying to fake an image could do much better with a train simulator.

 

[Sun Chariot]

Has anyone considered what this is going to do for photography? .... How will we then know that, say, when someone claims to have photographed a Class ??? at X they are telling the truth.

The power of "big data", already gives us a near-as real time insight of what rail traffic is where. We have means to prove, or disprove, any claim of "I saw..."
I think you allude to it in your post.

 

[Uncle Buck]

My wife says he looks like José Garcia, the French (Spanish parents) film star. There is something very weird looking out of the right hand train, just behind the cab.

Whereas the chap on the left hand train looking out the window, almost looking at the “camera”, looks like Ralph Cifaretto of The Sopranos, wearing dungarees- presumably the real reason for his mysterious disappearance was that he had a passion for suburban EMUs.

 

[GusB]

The text I used in Gemini was:

please generate an image of a scotrail class 170 in british rail blue and grey livery

The background looks very "West Highland", which would probably account for the 156-ish look. I don't think the result is too bad, but it's not a 170!

 

[The exile]

The power of "big data", already gives us a near-as real time insight of what rail traffic is where. We have means to prove, or disprove, any claim of "I saw..."
I think you allude to it in your post.

Though it will presumably create a field day for would-be fakers of historic images. No record of whether loco xyz actually worked to Little Whinging and harder to “fact check” the background.

 

[frankmoh]

Though it will presumably create a field day for would-be fakers of historic images. No record of whether loco xyz actually worked to Little Whinging and harder to “fact check” the background.

I remember the day fondly when the APT-P stopped at Addiscombe back in 1982...

 

[Sun Chariot]

Though it will presumably create a field day for would-be fakers of historic images. No record of whether loco xyz actually worked to Little Whinging and harder to “fact check” the background.

With a data source like TRUST and so many other sites etc to corroborate the data, my view is that it'll be relatively simple to snare the 'fake' images.

There is some good historic 'gen', such as Six Bells Junction and Derby Sulzers, so I think we so have good ways to check the suspicious images.

Historic images, I think we will all take with a pinch of salt if they 'magically' only come to light in the AI era. I for one will be highly sceptical.

 

[Sad Sprinter]

View attachment 183535

View attachment 183538

Feel free to suggest prompts (I have Gemini Pro) I might the more interesting some later

One of the things I love is how British these trains look. It’s really captured the feel of pre-privatised railways in Britain. This one is my absoloute favourite. I can see it as some medium-distance unit working Paddington to Westbury runs or something. The other day I tried to get ChatGPT to put the Swallow livery over a Shinkansen 300 set, but it just drew a class 91 instead.

 

[MarkyT]

That's an excellent representation of a class 108 DMU....apart from the fact that the fumes are coming out from the wrong place (about halfway along the side of the car!).

I presume there's a healthy wood burning stove, offering an at-seat service of wood fired pizza.

No, there's a smoky steam engine running light hiding behind the DMU on the adjacent track. The photographer came out specially to see it and the unit just appeared at the wrong moment!

 

[Sun Chariot]

No, there's a smoky steam engine running light hiding behind the DMU on the adjacent track. The photographer came out specially to see it and the unit just appeared at the wrong moment!

brilliant!

== Doublepost prevention - post automatically merged: 12 Jul 2025 ==

I couldn't resist trying this for myself. Here are....
1) A High Speed Train passing through lush green countryside.
2) A Merry Go Round Train passing a coal fired power station.
3) A Eurostar speeding through the Channel Tunnel, on fire.
4) A passenger train with people dancing on the roof and leaning out of the windows.
5) A SNCF TGV going fast in the snow past a beautiful sunset.

Now, which ones do you think will foul the platforms, signals and bridges?

 

[Sad Sprinter]

Following the success of the Intercity 125 rollout on the Western region, BR began to look at building a high speed line to speed up London to Glasgow journeys. Here an early 90s build Intercity 300 EMU glides past the Colne Valley reservoirs on its way north in October 1998. The Japanese influence on Metro-Cammell’s first high speed train design is clearly evident.

 

[Sun Chariot]

However, going completely OT, this Gemini AI image reminds me of my student days...

 

[duffield]

I hope you're all proud of the fact that you're adding more "AI slop" to the internet, and in particular that some of these fanciful creations will probably appear as the results of image searches by people who would really prefer the genuine article!


(Yes, I realise that this is a tiny, tiny insignificant drip in the vast ocean of AI slop that's being added to the internet every day).

 

[GusB]

2) A Merry Go Round Train passing a coal fired power station.

No gauging issues with that one!

 

[Class15]

I hope you're all proud of the fact that you're adding more "AI slop" to the internet, and in particular that some of these fanciful creations will probably appear as the results of image searches by people who would really prefer the genuine article!

Why is all AI “slop”?

 

[frankmoh]

Why is all AI “slop”?

Slop is a shortening for "sloppy content" since many AI things are poorly made.

 

[61653 HTAFC]

I hope you're all proud of the fact that you're adding more "AI slop" to the internet, and in particular that some of these fanciful creations will probably appear as the results of image searches by people who would really prefer the genuine article!


(Yes, I realise that this is a tiny, tiny insignificant drip in the vast ocean of AI slop that's being added to the internet every day).

To be fair, if anything this thread is highlighting the flaws of AI-generated content.

Surprised nobody has pointed out (unless I missed it) how much the effort in Post #1 looks like it was built by Metro-Cammell.

 

[duffield]

Why is all AI “slop”?

I didn't say "all AI is slop". The vast majority of AI generated content that's being added to the user posted part of the public internet daily seems to be though.

"AI slop" in my view is anything generated by AI which dilutes and displaces genuine original content, particularly by burying it in the search results.

For example, if you are searching for a real picture of a genuine class xxx unit and get instead a fantasy unit which never existed, that's almost certainly not what you want and will degrade your search experience or if it's convincing enough, will positively mislead you (into - say - thinking that a particular class was ever painted in a particular livery, or had certain types of windows)

When I search for technical articles on Linux, and increasingly get as my top results multiple obviously AI generated pages ripping off the original sources, that's "AI slop". You can read several paragraphs and suddenly find it doesn't make sense anymore, it's time wasting and frustrating. Sometimes the advice is stitched together such that it appears up to date but is dangerously outdated and will cause data loss, for example. Newly generated "AI slop" will appear up to date when doing search by time period even if it's based on ancient obsolete training data.

I've read that Google is struggling with an overwhelming tide of badly made unoriginal AI generated videos being added to Youtube every day (in the hope that some of it will reach the monetisation thresholds), making it harder to find original content.


The problem existed before the current AI boom, cheaply written rip-off content designed to attract clicks was increasing visible. But AI means that you can churn out literally thousands of articles, videos, songs, pictures etc. and flood the parts of the internet that allow user content with them.

 

[Class15]

I didn't say "all AI is slop". The vast majority of AI generated content that's being added to the user posted part of the public internet daily seems to be though.

"AI slop" in my view is anything generated by AI which dilutes and displaces genuine original content, particularly by burying it in the search results.

For example, if you are searching for a real picture of a genuine class xxx unit and get instead a fantasy unit which never existed, that's almost certainly not what you want and will degrade your search experience or if it's convincing enough, will positively mislead you (into - say - thinking that a particular class was ever painted in a particular livery, or had certain types of windows)

When I search for technical articles on Linux, and increasingly get as my top results multiple obviously AI generated pages ripping off the original sources, that's "AI slop". You can read several paragraphs and suddenly find it doesn't make sense anymore, it's time wasting and frustrating. Sometimes the advice is stitched together such that it appears up to date but is dangerously outdated and will cause data loss, for example. Newly generated "AI slop" will appear up to date when doing search by time period even if it's based on ancient obsolete training data.

I've read that Google is struggling with an overwhelming tide of badly made unoriginal AI generated videos being added to Youtube every day (in the hope that some of it will reach the monetisation thresholds), making it harder to find original content.


The problem existed before the current AI boom, cheaply written rip-off content designed to attract clicks was increasing visible. But AI means that you can churn out literally thousands of articles, videos, songs, pictures etc. and flood the parts of the internet that allow user content with them.

Fair enough, what I would say is that AI is relatively good at making the backscene and scenery nice in images though.

 

[Sun Chariot]

Interestingly, Google Gemini image creation seems to render SNCF TGV units far better than its "British trains". It even gets convincing catenary (even if their pantographs are away from the wires!); and that Thalys TGV seems to be formed entirely of power cars!

 

[Class15]

I asked it for a Freightliner Class 90, and I got, well, something similar to what I usually get when I want to see an actual Freightliner Class 90 on a container train:

Image shows AI blue class 66